UTF-8 is self-resynchronizing. You can scan forwards and/or backwards and all you have to do is look for bytes that start a UTF-8 codepoint encoding to find the boundaries between codepoints. It's genius.